I was pleased to be able to say to myself that what I do is exactly what they suggest you should do which is to drop down a level in automation when things get complicated.
But that is because I don't have much choice
I've got STARs in the GPS (not the newer RNAV ones) but I almost never load them because in almost every case I know of I was vectored anyway.
I also wonder how a runway change would be handled in practice, in Europe, in a non vectored environment.
If it occurs
before the enroute termination (i.e. before you reached the start of the STAR) then you can just load the new STAR. The autopilot will be in NAV mode and nothing changes.
If it occurs
after the enroute termination then what the hell are you supposed to do? ATC must consider this, because they will be faced with traffic departing one STAR part way through and intercepting the new one. If you (or your GPS) are dumb in doing this, you will end up doing something like a 180 because when you load the new STAR, it will get inserted into the route
whole i.e. you will now be taken back to its starting waypoint
I've never heard of a defined procedure for doing such a thing.
In nice wx you would deal with a runway change by asking for a visual approach and just hand fly to the other runway, or fly a circle to land. In OVC00x you can't just do that, and this is where overly-general comments run out of steam. At simplest, if too close in, you have to go around and do it all again.
I am not sure I agree with what I think he is suggesting which is to drop out the autopilot when you don't know what it is doing. Far better to leave it engaged but set the heading bug to where you want to go and then press HDG. Most pilots don't fly on instruments all that well when they have been suddenly dropped into it.